Constructing an accurate health monitoring (DHM) model plays a vital role in the operation cycle of crucial infrastructure such as dam. However, the dam health monitoring model with many input parameters would lead to unproper interpretation of the complex relationship between input variables and response. Considering the comprehensive influence of various factors on dam effect quantities such as deformation and seepage, a dimensionality reduction modeling method of dam safety monitoring model based on the combination of least squares support vector machine (LSSVM) and manifold learning is proposed. Combined with the particle swarm optimization (PSO) algorithm, the locality preserving projections (LPP) technology in manifold learning and the support vector machine (SVM) method, the optimization of input factors and the determination of model parameters of SVM model are explored, to obtain the nonlinear mapping relationship between the effect quantity and its influencing factors. The application of an engineering example using the prototype monitoring data resources shows that the proposed method, compared with the traditional statistical model or the combination of the above models, the model constructed by the proposed method in this paper can effectively reduce the workload and improve the prediction accuracy has certain advantages in improving the modeling efficiency and prediction accuracy.